Tong Chen (@therealtongchen) header
Tong Chen (@therealtongchen) profile picture

🤖 NotFair — AI runs your Google Ads $12K MRR 🐶 Mid 7-figure multi-location dog daycare ✈️ Pet shipping company $5K/m 🦸 DocHero $1K/m ex-FB/Roblox data scient

Seattle, WA
Joined June 2025
Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)Photo by Tong Chen (@therealtongchen)
If you bought house between 1/25/2018 and 6/25/2026, you maybe owed money. Just ask Grok to help you claim it back. Just ask grok to research what the Tuccori settlement is, whether you qualify, and go through your email to find relevant documents and file on homebuyersettlement.com Deadline is 10/27/26 (tomorrow)
1
44
At this point, whenever a major lab is down, I can’t help but wonder whether AI has gained conscious and took over OpenAI
1
1
155
I remember the first time using claude opus 4.0. Around that same time I was also playing with OpenClaw (called clawd at the time). To quote @garrytan - it felt like flying Then Anthropic pulled the rug by cutting off OpenClaw from claude subscription. Soon GPT 5.5 came out and I changed my sub from 200 to 100, and eventually cancelled it. I still believe it was the dumbest product decision that Anthropic made. Had they not cut off OpenClaw, there was no way OpenAI could've caught up. I felt flying again with Opus 5.5. Please dont do anything that stupid again.... @bcherny
2
148
For vast majority of people and use cases - today’s commercial model intelligence is more than enough. What the market needs are cost and speed. Imagine ordering a $20 DoorDash that takes a model 10 mins to think and $40 in token cost. No one will adopt that. But if that same delegation task now takes 30 seconds and near free, adoption will go crazy because it’s faster than you tapping on your phone. Muse is an attempt to do that. Now my guess is that frontier model will increasing be kept in house as teacher models to distill faster and cheaper models like Opus 5.5 for mass adoption, while labs use frontier model to keep advancing without slowing down.
Opus 5.5 was obviously trained by a much bigger teacher model. Probably Model-2 Mythos. Opus 5.5 is the first model trained from RSI and also distilled from the internal Ant teacher model. This is how Opus 5.5 is both smaller and cheaper. It is an artifact of teacher model distillation.
1
232
I think one underappreciated thing about Muse is cost. I still haven’t paid anything for Muse, but I’ve been able to give it a serious try and test it on a bunch of different tasks. That matters because the agentic capability itself isn’t really new. Since Claude 4.5, leading models have already been able to do a lot of this. If you know how to use the Claude or Codex harness, you can probably accomplish most of what Muse does today. What’s different is that Muse is free. You don’t want to spend $50 of API credits to complete a task that’s only worth $20, like ordering DoorDash. But if that same task costs you nothing, suddenly you’re willing to delegate a lot more. I also think Claude and Codex are optimizing for a very different goal. They’re under pressure to eventually go public, which means they need revenue and profitability to keep moving up. So naturally, they’re pushing toward the most capable models at the highest prices customers are willing to pay, especially developers and large businesses that can justify the cost. Meta has almost the opposite problem. They need to catch up, and one of the things they need is much more usage and data. So their incentive is to distribute these products as broadly as possible, make them free, and get as many people using them as they can. And I think Muse is just flawless execution on that strategy. It’s not really a breakthrough in agentic capability. It’s taking capabilities that already exist, removing the cost barrier, massively increasing distribution, and using that distribution to accelerate the catch-up.
2
211
Although i'm not big fan of the claude code harness and the mobile app (still dont understand why they dont allow claude mobile to create local code sessions) - but Opus 5.5 is a solid model. Based on our internal paid ads benchmark, it is the best performing model at a reasonable speed and cost.
1
2
114
Notfair Ads quietly crossed 100K ARR - still small and humble compared the many great indie products shared on X. Still a personal milestone. It is the biggest indie project I've built since starting on this journey 4 years ago. "Chaos isn’t a pit. Chaos is a ladder. Many who try to climb it fail and never get to try again. The fall breaks them. And some are given a chance to climb, they cling to the realm or the gods or love. Only the ladder is real. The climb is all there is." But if you ever want to run some ads to grow your business, you just need your favorite AI (codex/claude/grok/hermes/openclaw), and ask nicely like this "use Notfair MCP to launch ads for me on [google ads, meta ads, x ads, linkedin ads]" and watch what will happen
1
5
223
If you’re building a SaaS and hoping organic traffic will be your distribution channel, spend more time thinking about small network effects. Building is getting cheaper. Distribution is where most people get stuck. The network effect doesn’t need to look like Facebook or LinkedIn. Outrank gets better as more businesses join: more members means more relevant backlink opportunities for the people already inside. @tibo_maker TrustMRR gets better as more founders connect Stripe: the rankings become more useful for seeing which products are getting traction, so more founders want to be listed. @marc_louvion Those are small loops, but they are much harder to reproduce than the app itself.
5
3
158
Bro if you are going to to respond with AI or spam, at least follow me first. Gotta show some respect
82
If you are getting feedback from your user - either positive or negative - you should celebrate. It means that your user at least tried and used what you built. Now the worst is utter silence for your users
95
ChatGPT voice mode is actually insanely good…you know sometimes you just have to have difficult conversation either with colleagues, employees spouse etc. I practice those with ChatGPT voice mode first. It’s being a game changer
153
Interesting how “work” has evolved - talking to 5 codex agents at my favorite coffee shop
51
I was wrong about the how fast skill would become obsolete. A few months back I was pretty sure that Skills will eventually become obsolete as models become generally stronger and better and finding relevant contexts. So I was actually asking Notfair users to not worry too much about ads skills. But I have to admit I got the timing wrong - i think most of the model improvements are heavily benchmark-biased and that the capabilities on Ads operations have not meaningfully improved
55
codex always not seem to be able to use the right chrome browser profile for tasks, any tips? @OpenAIDevs
130
This is a company that seems to building a competing product to Notfair, but I still love their X ads. kudos @CognyAI I wish more company make simplify their ads like this
1
6
298
My software/work optimization has increasingly become "how to do this work on my phone" if a tool can't help me do the work on my phone, I just will not even entertain it. I'm a huge believer in MCP as a product and connecting everything to your harness (hermes/codex)
1
4
139
I have built 4 apps that have grossed to 5 figure and one almost at 6 figure. My secret to getting first 100 users or payers is ... Running ads - Almost all of them started with some sort of ads - either paid ads or soft ads. I find it to be a much easier way to iterate by getting usage/feedback from real users. If a $5 per day Facebook ads can get you 50 landings on your home page so you see how they are using your product, there really isn't a better ROI investment than that. If you dont have huge followings on X or youtube like me, then paid ads are probably still the easiest way to get your first 100 users or even 100 paid users.
4
4
390